Add to Chrome
Add to Firefox
Add to EdgeACFM: A Dynamic Spatial-Temporal Network for Traffic Prediction
Sep 02, 2019
As a crucial component in intelligent transportation systems, crowd flow prediction has recently attracted widespread research interest in the field of artificial intelligence (AI) with the increasing availability of large-scale traffic mobility data. Its key challenge lies in how to integrate diverse factors (such as temporal laws and spatial dependencies) to infer the evolution trend of crowd flow. To address this problem, we propose a unified neural network called Attentive Crowd Flow Machine (ACFM), which can effectively learn the spatial-temporal feature representations of crowd flow with an attention mechanism. In particular, our ACFM is composed of two progressive ConvLSTM units connected with a convolutional layer. Specifically, the first LSTM unit takes normal crowd flow features as input and generates a hidden state at each time-step, which is further fed into the connected convolutional layer for spatial attention map inference. The second LSTM unit aims at learning the dynamic spatial-temporal representations from the attentionally weighted crowd flow features. Further, we develop two deep frameworks based on ACFM to predict citywide short-term/long-term crowd flow by adaptively incorporating the sequential and periodic data as well as other external influences. Extensive experiments on two standard benchmarks well demonstrate the superiority of the proposed method for crowd flow prediction. Moreover, to verify the generalization of our method, we also apply the customized framework to forecast the passenger pickup/dropoff demands and show its superior performance in this traffic prediction task.
